You are on page 1of 33

Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu

Applied Delayed Dierential Equations


Chang-Yuan Cheng
National Pingtung University of Education
cycheng@mail.npue.edu.tw
NCTS Summer Course on Dynamical Systems, July 17,19, 2012
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Outline
1 Introduction
2 Delayed Negative Feedback
3 Existence of Solutions
4 Linear Systems and Linearization
5 Examples and Lyapunov Functions/Functionals
6 Hopf Bifurcation
Exercise :Five take-home problems. Due: Aug. 24.
(cycheng@mail.npue.edu.tw)
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
References
H. Smith, An Introduction to Delay Dierential Equations with
Applications to the Life Sciences, 2011
Y. Kuang, Delay Dierential Equations with Applications in Population
Dynamics, Academic Press, New York, 1993.
J. Mallet-Paret and G. Sell, The Poincare-Bendixson theorem for
monotone cyclic feedback systems with delay, J. Di. Eqns. 125 (1996),
441-489.
L. Shayer and S. A. Campbell, Stability, bifurcation, and multistability in a
system of two couplrd neurons with multiple time delays, SIAM AM. 61
(2000), 673-700.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Examples
The logistic equation describing the growth of a single population
N

(t) = N(t)[b aN(t)].


Population density negatively aects the per capita growth rate,
environmental degradation.
Hutchinson: time delay due to developmental and maturation delays.
N

(t) = N(t)[b aN(t r)]


(constant delay)(discrete delay), where a, b, r > 0, r is called the delay.
Wrights conjecture remains open.
N

(t) = N(t)[b a
_

0
N(t s)k(s)ds]
(distributed delay).
Normalized,
_

0
k(s)ds = 1.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Examples
Nicholsons blowy equation:
N

(t) = bN(t r)exp(


N(t r)
N
0
) N(t).
Blowy eggs, larvae, sexually mature ies, for adult ies eggs take exactly
r time units to develop into adults.
The rst term: accounts for recruitment of new adults, a product of a
fecundity term and a probability of survival from egg to adult.
The nal term: death.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Examples
Busenberg and Cooke
y

(t) = b(t)y(t T)[1 y(t)] cy(t).


After a delay T during which the infectious agent develops. Seasonal
periodic incidence: 0 b(t) = b(t +). Example:
b(t) = b(1 +a sin(2t/)), 0 < a < 1. c is the recovery rate.
A single self-excited neuron with delayed excitation:
x

(t) = x(t) +tanh(x(t )).


x(t) encodes the neurons activity level. The delay represents the
transmission time between output x(t) and input.
y

i
(t) = A
i
y
i
(t) +
n

j=1
W
ij
tanh(y
j
(t
ij
)) + I
i
(t).

ij
is the delay in signal transmission between the jth neuron and ith
neuron.
W i d R h th t ill ti f th t
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Examples
x(t r): a discrete delay
_
t
tr
k(k s)x(s)ds =
_
r
0
k(z)x(t z)dz,
where 0 r , a distributed delay.
If the delay kernel k(u) is identically zero for all u > u
max
, is said to be a
bounded delay.
Otherwise, we call the delay an unbounded delay or an innite delay.
Delay may be time-dependent such as x(t r(t)) where r(t) 0. It may
also be a state-dependent delay such as s(t r(x(t))).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Solving DEs Using a Computer
A tutorial on solving delay equations using MATLAB by Shampine
and Thompson can be found at the web site
http://www.runet.edu/ thompson/webddes/
A list of software available for delay dierential equations is
maintained at the site
http://twr.cs.kuleuven.be/research/software/delay/software.shtml
Among these is DDEVBIFTOOL v. 2.03, a MATLAB package for
bifurcation analysis of delay dierential equations.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Preliminaries
u

(t) = u(t ). (1)


The method of steps.
By using the MATLAB DDE23 package.
Fig.: Solution of equation (1) with initial data u() = 1, [1, 0], for
various .
= 0.25-no oscillations
= 0.6-hint of oscillations
= 1.0-damped oscillations
= 1.5-barely damped oscillations
= 2.0-undamped oscillations
u 0 is a solution, a steady-state solution. It is stable when < /2
and unstable when > /2 1.58.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
u

(t) = u(t r), (2)


r 0. The case > 0 is of most interest inasmuch as it corresponds to
negative feedback; > 0 is the positive feedback case. When r = 0,
u = 0 is an asymptotically stable steady state for the case of negative
feedback; it is unstable for positive feedback. What happens when r > 0?
U() = u(t). =
t
r
. = r.
dU
d
= U( 1). (3)
We seek (complex) values of such that U() = exp() is a solution of
(3). It is convenient to introduce the linear operator, dened on the
dierentiable functions, by
L(U) =
dU
d
+U( 1).
Then
L(e

) = e

+e
(1)
= e

[ +e

].
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
L(e

) 0 (the zero function) if and only if is a root of the


characteristic equation
h() +e

= 0. (4)
We call C a root of (4) of order l, where l 1, if
h() = h

() = h

() = = h
(l1)
() = 0, h
(l)
() ,= 0.
Lemma

j
e

, j = 0, 1, , k are solutions of (3) if and only if is a root of


order at least k + 1 of h.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
Alternatively, (4) can be obtained by using the Laplace transform.
Based on our experience with ODEs, we have a right to expect that the
trivial solution is asymptotically stable if Re() < 0 for all roots of the
characteristic equation and that it is unstable if there is a root with
positive real part.
Letting = x +iy and considering real and imaginary parts, we get
x = e
x
cos(y)
y = e
x
sin(y).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
Lemma
The following hold.
If < 0, then there is exactly one real root and it is positive.
If 0 < < e
1
, then there are exactly two real roots x
1
< x
2
, both
negative. x
1
and x
2
0 as 0.
If = e
1
, then there is a single real root of order two, namely
= 1
If > e
1
, then there are no real roots.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
Proposition
The following hold for (4).
1. If 0 < < /2, then there exists > 0 such that R() for
all roots.
2. If = /2, then = i/2 are roots of order one.
3. If > /2, there are roots = x iy with x > 0, y (/2, ).
Corollary
The following hold for (2).
If < 0, then u = 0 is unstable.
If 0 < r < /2, u = 0 is asymptotically stable.
If r = /2, u = sin(/2), cos(/2) are solutions.
If r > /2, u = 0 is unstable.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Simplest DE
Figure: The stability region in the (r, )-plane for (2).
Simulation: Simulation of (2) for = 1, r = 1.4 and 1.58, I.C. = 1.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Oscillation of Solutions
Theorem
For every real and r > 0 the following are equivalent.
Every solution of (2) is oscillatory.
r > 1/e.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Method of Steps
Existence and uniqueness of solutions of discrete-delay dierential
equations is established by the method of steps, appealing to classical
ODE results.
More general delay equations require a more general framework for
existence and uniqueness.
Solutions either extend to the entire half-line or blowup in nite time.
Applications to biology require that solutions that start positive, stay
positive in the future.
Dierential inequalities involving delays are important tools with which to
bound solutions.
x

(t) = f(t, x(t), x(t r)) (5)


x(t) = (t), s r t s (6)
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Method of Steps
Theorem
Let f(t, x, y) and f
x
(t, x, y) be continuous on R
3
, s R, and let
: [s r, s] R be continuous. Then there exist > s and a unique
solution of the initial-value problem (5) on [s r, ].
The previous theorem provides only a local solution of (5). Just as for
ODEs, we can often, but not always, extend this solution to be dened
for all t s.
Theorem
Let f satisfy the hypotheses of The previous theorem and let
x : [s r, ) R be the noncontinuable solution of the initial-value
problem (5). If < then
lim
t
[x(t)[ =
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Positivity of Solutions
Most delay dierential equations that arise in population dynamics and
epidemiology model intrinsically nonnegative quantities. It is important
to establish that nonnegative initial data give rise to nonnegative
solutions. Consider
x

(t) = y(t) x(t r)


y

(t) = y(t)
If r = 0, then solutions corresponding to nonnegative initial data remain
nonnegative in the future.
By applying the variation of constants formula to the rst equation
x(t) = x(0)e
t
+e
t
_
t
0
e
s
y(s)ds.
If r > 0, it fails.
Take y(0) = 0 and x(s) = s/r, r s 0.
x(0) = 0 and x

(0) = 1.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Positivity of Solutions
The solution immediately becomes negative yet the initial data are
nonnegative.
x 0 when x
i
0, 1 i n; R
n
+
denotes the set of vectors x R
n
such that x 0.
Theorem
Suppose that f : R R
n
+
R
n
+
R
n
satises that f(t, x, y) and
f
x
(t, x, y) are continuous on R
3
, and
i, t, x, y R
n
+
: x
i
= 0 f
i
(t, x, y) 0.
If the initial data in (5) satisfy 0, then the corresponding solution
x(t) of (5) satises x(t) 0 for all t s where it is dened.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Positivity of Solutions
Example: the discrete-delay predator-prey model
N

1
(t) = N
1
(t)[b N
2
(t)] (7)
N

2
(t) = N
2
(t)[c +N
1
(t r)],
where b,c > 0 and delay r > 0 reects a delay in assimilation of
consumed prey. Nonnegative initial conditions:
N
1
(t) =
1
(t), r t 0 (8)
N
2
(0) = N
0
2
.
x = (x
1
, x
2
) where x
i
= N
i
(t), y = (y
1
, y
2
) where y
i
= N
i
(t r)
f(t, x, y) = (x
1
[b x
2
], x
2
[c +y
1
])
f,f
x
exist and are continuous so it has a unique noncontinuable solution
dened on some interval [r, ) where > 0.
This solution has nonnegative components.
If x,y 0 and x
i
= 0 for some i, then f
i
(t, x, y) = 0, = +.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Stability Denitions
Consider
x

(t) = f(t, x
t
).
Suppose f(t, 0) = 0, t R.
The solution x = 0 is stable if for any R and > 0, there exists
= (, ) > 0 such that c and || < implies that
|x
t
(, )| < , t .
It is asymptotically stable if it is stable and if there exists b() > 0 such
that whenever C and || < b(), then x(t, , ) 0, t .
x = 0 is unstable if it is not stable.
The stability of any other solution can be dened by changing variables
such that the given solution is the zero solution.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Exercise
Exercise 1: Aiello and Freedman introduce a model of a stage-structured
population consisting of immature x
1
and mature x
2
individuals:
x

1
(t) = rx
2
(t) dx
1
(t) e
d
x
2
(t )
x

2
(t) = e
d
x
2
(t ) ax
2
2
(t).
Do nonnegative initial data give rise to nonnegative solution?
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Automous Linear Systems
Although the method is similar to that for ODEs, the characteristic
equation is more complicated, typically having innitely many roots.
Fortunately, all but nitely many of these roots have real part less than
any given real number.
In this chapter, c = C([r, 0], C
n
).
A function L:c C
n
is linear if
L(a +b) = aL() +bL(), , c, a, b C.
L is said to be bounded if there exists K > 0 such that
[L()[ K||, C
x

(t) = L(x
t
) (9)
Let A and B be n n matrices and dene
L() = A(0) +B(r).
Then
[L()[ [A[[(0)[ +[B[[(r)[ ([A[ +[B[)||
L is bounded.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Automous Linear Systems
x

(t) = Ax(t) +Bx(t r)


x

(t) = Ax(t) +
m

j=1
B
j
x(t r
j
)
Assume that r
i,j
> 0, i, j = 1, 2. k
ij
: [0, r
ij
] C.
x

1
(t) =
_
r
11
0
k
11
(s)x
1
(t s)ds +
_
r
12
0
k
12
(s)x
2
(t s)ds (10)
x

2
(t) =
_
r
21
0
k
21
(s)x
1
(t s)ds +
_
r
22
0
k
22
(s)x
2
(t s)ds
Let r = max r
ij
and extend k
ij
to [0,r] if necessary by making it
identically zero on (r
ij
,r]. k(s) = (k
ij
(s)), x(t) = (x
1
(t), x
2
(t)).
Then
x

(t) =
_
r
0
k(s)x(t s)ds
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Laplace Transform and Variation of Constants Formula
x

(t) = Ax(t) +Bx(t r) +f(t), t 0, x


0
= , (11)
where A, B are scalars; the case that they are matrices is treated by
analogy.
The Laplace transform:
F(s) = L(f) =
_

0
e
st
f(t)dt,
where f : [0, ) C is an exponentially bounded function,
[f(t)[ Me
kt
for some real M, k.
If f, g:[0, ) C then their convolution is dened by
(f g)(t) =
_
t
0
f()g(t r)d =
_
t
0
f(t )g()d
Then
L(f g) = F(s)G(s),
where F = L(f) and G = L(g).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Laplace Transform and Variation of Constants Formula
Applying the transform to (11)
sX(s) (0) = AX(s) +B[
_
r
0
e
st
(t r)dt +
_

r
e
st
x(t r)dt] +F(s
= AX(s) +B[
_
r
0
e
st
(t r)dt +e
sr
X(s)] +F(s)
= [A+e
sr
B]X +B
_
r
0
e
st
(t r)dt +F(s)
= [A+e
sr
B]X +B(s) +F(s), (12
= L(( r)) and extending to [r, ) by making it zero for t > 0.
X(s) = K(s)[(0) +B(s) +F(s)]
where
K(s) = (s A e
sr
B)
1
k: the inverse transform of K. k is the solution of (11), with f=0, for
the initial data
() =
_
1, = 0
0, r < 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Laplace Transform and Variation of Constants Formula
In fact, k(t) = e
At
, 0 t < r, and it satises the initial value problem:
x

(t) = Ax(t) +Be


A(tr)
, x(r) = e
Ar
, r t < 2r.
k is called the fundamental solution of (11).
x(t) = x(t; , f) = x(t; , 0) +x(t; 0, f)
x(t; 0, f) =
_
t
0
k(t )f()d
x(t; , 0) = k(t)(0) +
_
t
0
k(t )B( r)d
These formulas hold with minor changes in the case that (11) is a vector
system with matrices A, B.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Characteristic Eq.
We seek exponentially growing solutions of (9),
x(t) = e
t
v, v ,= 0.
is complex, v is a vector whose components are complex on [r, 0],
exp

() = e

.
x
t
() = x(t +) = e
(t+)
v = e
t
exp

()v
x

(t) = e
t
v = L(x
t
) = e
t
L(exp

v)
v = L(exp

v)
Writing v =

j
v
j
e
j
where e
j

j
is the standard basis for C
n
, then
L(exp

v) =

j
v
j
L(exp

e
j
).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Characteristic Eq.
Dene the n n matrix
L

= (L(exp

e
1
)[L(exp

e
2
)[ [L(exp

e
n
)) = (L
i
(exp

e
j
)),
where L
i
() is component i of L().
Then L(exp

v) = L

v and x(t) = e
t
v is a nonzero solution of (9) if
is a solution of the characteristic eqution:
det(I L

) = 0.
Lemma
h() = det(I L

) is an entire function.
Properties of nontrivial entire functions, in particular of h, are:
Each characteristic root has nite order.
There are at most countably many characteristic roots.
The set of characteristic roots has no nite accumulation point.
Remarkably, there are only nitely many characteristic roots with positive
real part.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Characteristic Eq.
Lemma
Given R, there are at most nitely many characteristic roots
satisfying Re() > . If there are innitely many distinct characteristic
roots
n

n
, then
Re(
n
) , n
An important implication of the previous lemma is that there exists
R and a nite set of dominant characteristic roots having maximal
real part equal to with all other roots having real part strictly less than
.
Proposition
Suppose that L maps real functions to real vectors:
L(C([r, 0], R
n
)) R
n
. Then is a characteristic root if and only if

is a characteristic root.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Characteristic Eq.
Theorem
Suppose that Re() < for every characteristic root . Then there
exists K > 0 such that
[x(t, )[ Ke
t
||, t 0, c
where x(t, ) is the solution of (9) satisfying x
0
= . In particular,
x = 0 is asymptotically stable for (9) if Re() < 0 for every
characteristic root; it is unstable if there is a root satisfying Re() > 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Small Delays Are Harmless
The linear delay system
z

(t) = Az(t) +Bz(t r)


Nondelayed counterpart
z

(t) = (A +B)z(t)
the characteristic equation:
h(, r) = det[I A e
r
B] = 0 (13)
h(, 0) = det[I AB] = 0
Theorem
Let z
1
, z
2
, , z
k
be the distinct eigenvalues of A+B, let > 0, and let
s R satisfy s < min
i
Re(z
i
). Then there exists r
0
> 0 such that if
0 < r < r
0
and h(z, r) = 0 for some z then either Re(z) < s or
[z z
i
[ < for some i.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Small Delays Are Harmless
In words, for small enough delay, the characteristic roots of (13) are
either very near the eigenvalues of A +B or have more negative real
parts than any of the eigenvalues of A+B.
Small delays are harmless in the sense that if asymptotic stability holds
when = 0, then it continues to hold for small delays in as much as we
may choose small enough that the -ball about each eigenvalue of
A +B belongs to the left half-plane and we may choose s negative.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Scalar Equation
A and B are real scalars.
x

(t) = Ax(t) +Bx(t r) (14)


= A+Be
r
z = r, = Ar, = Br
z = +e
z
. (15)
If we write z = x +iy, then
0 = x e
x
cos(y)
0 = y +e
x
sin(y)
z = 0 is a root precisely when + = 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Scalar Equation
Dene F(z, , ) := z e
z
.
Setting x = 0 and solving for and gives the neutral stability curves
in parameter space
= y cos(y)/ sin(y)
= y/ sin(y)
We denote by
C
0
:= (, ) = (y cos(y)/ sin(y), y/ sin(y)), 0 y <
The curve along which z = iy, 0 y < are roots.
d
dy
< 0,
d
dy
< 0, 0 < y <
Searting from (1, 1) when y = 0, meets the -axis at (0, /2) when
y = /2.
Approaches (, ) from below and tangent to the line = as
y because / = cos(y) 1 whereas both , as
y .
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Scalar Equation
We also consider the curves
C
n
:= (, ) = (y cos(y)/ sin(y), y/ sin(y)), n < y < (n+1), n 1,
where z = iy, n < y < (n + 1) are roots.
On C
n
, d/dy < 0 but d/dy changes sign on (n, (n + 1)) where
tan(y) = y. [/[ > 1 on C
n
implying that C
1
, C
3
, lie strictly above
the graph of = [[ and C
2
, C
4
, lie strictly below the graph of
= [[.
Figure: Stability region for (15) in the (, )-plane lies to the left of the
displayed curve.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Scalar Equation
We assume that A +B ,= 0 for otherwise = 0 is a root.
Theorem
The following hold for (14)G
If A +B > 0, then x = 0 is unstable.
If A +B < 0 and B A, then x = 0 is asymptotically stable.
If A +B < 0 and B < A, then there exists r

> 0 such that x = 0


is asympototically stable for 0 < r < r

and unstable for r > r

.
In case (c), there exist a pair of purely imaginary roots at
r = r

=
cos
1
(A/B)

B
2
A
2
.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
Consider the nonlinear functional dierential equation
x

(t) = f(x
t
). (16)
Then x(t) = x
0
R
n
, r R is a steady-state solution of (16) if and
only if
f( x
0
) = 0
where x
0
c is the constant function equal to x
0
.
If x(t) is a solution of (16) and
x(t) = x
0
+y(t)
then
y

(t) = f( x
0
+y
t
).
To understand the behavior of solutions of y

(t) = f( x
0
+y
t
) for
solutions that start near y = 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
Assume that
f( x
0
+ ) = L() +g(), C,
where L : c R
n
is a bounded linear function and g : c R
n
is
higher order in the sense that
lim
0
[g()[
||
= 0.
The linear system
z

(t) = L(z
t
)
is called the linearized (or variational) equation about the equilibrium x
0
.
We must view it on the complex space c = C([r, 0], C
n
).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
Theorem
Let () = 0 denote the characteristic equation corresponding to
z

(t) = L(z
t
) and suppose that
:= max
()=0
Re() < 0.
Then x
0
is a locally asymptotically stable steady state of (16). In fact,
there exists b > 0 such that
| x
0
| < b |x
t
() x
0
| K| x
0
|e
t/2
, t 0.
If Re() > 0 for some characteristic root, then x
0
is unstable.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
Consider
x

(t) = F(x(t), x(t r))


where F : DD R
n
is continuously dierentiable and D R
n
is
open. If F(x
0
, x
0
) = 0 for some x
0
D, then x(t) = x
0
, t R is an
equilibrium solution. Then f() = F((0), (r)) so
f( x
0
+) = A(0) +B(r) +G((0), (r))
where A = f
x
(x
0
, x
0
), B = f
y
(x
0
, x
0
).
It follows that the linearized system about x = x
0
is
x

(t) = Ax(t) +Bx(t r)


Consider the scaled version of the delayed chemostat model
S

(t) = 1 S(t) f(S(t))x(t) (17)


x

(t) = e
r
f(S(t r))x(t r) x(t),
where f(S) = mS/(a +S) and x, S 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
Nonnegative equilibria consist of the washout state (S, x) = (1, 0) and,
if f(1) > e
r
, the survival state (S, x) = (

S, x), where f(

S) = e
r
and
x = (1

S)e
r
.
A =
_
1 xf

(S) f(S)
0 1
_
, B = e
r
_
0 0
f

(S)x f(S)
_
( + 1)( + 1 +xf

(S) e
r
f(S)e
r
) = 0
What luck that if factors! = 1, and
= 1 xf

(S) +e
r
f(S)e
r
For the washout state, a = 1 and b = e
r
f(1). Hence,
a +b = 1 +e
r
f(1) and b > a.
The washout state is asymptotically stable when e
r
f(1) < 1 but
unstable if e
r
f(1) > 1, when the survival state exists.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Linearized Stability
For the survival state, a = 1 xf

S) and b = 1, so a +b < 0 and


b > a. Therefore, the survival state is asymptotically stable when it exists.
Exercise 2: Show that z = 0 is a double root of z = +e
z
only when
= 1 and = 1. In fact, the only double roots are at z = 1 when
= 1 + log [[, < 0. There are no roots of order three or higher.
Exercise 3: Consider the Lotka-Volterra competition system
x

(t) = x(t)[2 ax(t) by(t r)]


y

(t) = y(t)[2 cx(t r) dy(t)]


Determine the stability of the positive steady state when a = d = 2
and b = c = 1. How does it depend on r?
Determine the stability of the positive steady state when a = d = 1
and b = c = 2. How does it depend on r?
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Logistic Equation
[Hutchinson, 1948], the delayed logistic equation
n

(t) = a[1 n(t T)/K]n(t).


Let N(t) = n(t)/K and rescale time, then
N

(t) = N(t)[1 N(t r)], t 0. (18)


We are primarily interested in nonnegative solutions.
Proposition
Every orbit of (18) with 0 is bounded. In fact, for each such ,
there exists T > 0 such that
0 N(t, ) e
r
, r > T
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Logistic Equation
Putting u = N 1, it becomes the famous Wrights equation
u

(t) = u(t r)[1 +u(t)]


The steady-state N = 1 is now u = 0, the linear equation
v

(t) = v(t r)
will determine the stability of our N = 1 steady state. Recall that if
r < /2 then v = 0 is asymptotically stable and if r > /2, then v = 0 is
unstable. Wright proved the following.
Theorem
(see [Kuang, 1993]) If r 3/2, then N(t, ) 1, t for all
solutions of (18) satisfying (0) > 0.
Note that 3/2 = 1.5 < 1.57 = /2. Wrights conjecture, still
unsolved, is that the previous theorem holds with /2 instead of 3/2.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Logistic Equation
Simulations: Simulations of (18) for dierent values of r
(r = 0.25, 1.52, 1.60); N
0
= 0.5.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Liapunov Functions
Consider
x

(t) = f(x
t
),
where f : C R
n
is locally Lipschitz and completely continuous.
Let x(t) = x(t, ) be the solution satisfying x
0
= .
Given V : C R, dene

V () = lim
h0+
1
h
[V (x
h
()) V ()]
if the limit exists.
Theorem
(The LaSalle invariance principle) If V is a Liapunov function on G and
x
t
() is a bounded solution such that x
t
() G, t 0, then () ,= 0 is
contained in the maximal invariant subset of S

G :

V () = 0.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Logistic Eq., Instantaneous/Delayed Density Dependence
Consider
x

(t) = rx(t)[1 a
1
x(t) a
2
x(t )]. (19)
We rewrite it as
x

(t) = rx(t)[a
1
(x(t) x

) a
2
(x(t ) x

)]
Let G = C : 0, (0) > 0 and dene V : G R by
V () = (0) x

log((0)/x

) +
_
0

((s) x

)
2
ds
where > 0 is to be determined.
Exercise 4: Verify that V () satises:
V is continuous on G and becomes innite at a boundary point
where (0) = 0,
it is positive denite with respect to X

C, the function
identically equal to x

, in the sense that


V () > 0 = V (X

), G, ,= X

.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Logistic Eq., Instantaneous/Delayed Density Dependence
If x(t, ) is a solution of (19) with G, then x
t
() G for t 0 and
V (x
t
) = x(t) x

log(x(t)/x

) +
_
0

(x(t +s) x

)
2
ds
= x(t) x

log(x(t)/x

) +
_
t
t
(x(s) x

)
2
ds (20)
d
dt
V (x
t
) =
x(t) x

x(t)
x

(t) +[(x(t) x

)
2
(x(t ) x

)
2
]
=
x(t) x

x(t)
rx(t)[a
1
(x(t) x

) a
2
(x(t ) x

)]
+[(x(t) x

)
2
(x(t ) x

)
2
]
= ra
1
(x(t) x

)
2
ra
2
(x(t) x

)(x(t ) x

)
+[(x(t) x

)
2
(x(t ) x

)
2
]
= (ra
1
)(x(t) x

)
2
ra
2
(x(t) x

)(x(t ) x

)
(x(t ) x

)
2
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Logistic Eq., Instantaneous/Delayed Density Dependence
If we take = ra
1
/2, then
d
dt
V (x
t
) = (ra
1
/2)(x(t) x

)
2
ra
2
(x(t) x

)(x(t ) x

)
(ra
1
/2)(x(t ) x

)
2
= W(x(t) x

, x(t ) x

) (21)
where
W(u, v) = (ra
1
/2)u
2
ra
2
uv (ra
1
/2)v
2
=
r
2
_
a
1
a
2
a
2
a
1
__
u
v
_

_
u
v
_
(22)
The symmetric matrix has eigenvalues a
1
a
2
, both positive if a
1
> a
2
,
and therefore
d
dt
V (x
t
) = W(x(t) x

, x(t ) x

) 0, t 0, G
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Logistic Eq., Instantaneous/Delayed Density Dependence
Theorem
If a
1
> a
2
and G, then () = X

and x(t, ) x

as t .
Exercise 5: Cooke modeled a vector borne disease by
y

(t) = by(t r)[1 y(t)] cy(t)


where b, r > 0 and c 0. y(t) denotes the fraction of human population
that is infected with the disease, c is the recovery rate, r is the delay
between when a vector (e.g., a mosquito) becomes infected after biting
an infected person until the time when its bite can infect a human, and b
is the contact rate. Show that G = c : 0 () 1 is positively
invariant as the model demands. If c > b > 0, show that the trivial
solution is globally attracting. Hint: the Lyapunov function
V () =
1
2c
(0)
2
+
1
2
_
0
r

2
()d.
PS: You may also use the monotone dynamics.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Canonical Example
Consider
x

(t) = x(t 1)[v +x


2
(t) +x
2
(t 1)], (23)
where v is a real parameter. 0 is a steady state for all v,
_
v/2 is a
steady state for v > 0. Focus on the steady-state x = 0 rst. Linearizing
about x = 0, we get
y

(t) = vy(t 1)
Its characteristic equation is
= ve

y = 0 is asymptotically stable for 0 < v < /2. At v = 0, = 0 is the


only root. At v = /2, = i/2 are roots. Therefore x = 0 is
asymptotically stable for (23) when 0 < v < /2.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Canonical Example
Figure: Steady-steady bifurcation and Hopf bifurcation for (23).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Canonical Example
At v = /2, y(t) = sin(t/2), y(t) = cos(t/2). Try
x(t) = a sin(t/2)
a

2
= av a
3
x(t) =
_
v

2
sin(t/2), v >

2
is a periodic solution of (23).
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Hopf Bifurcation Theorem
x

(t) = F(x
t
, ) (24)
F : C R R
n
is twice continuously dierentiable
F(0, ) 0.
F(, ) = L() +f(, )
lim
0
[f(, )[
||
= 0
The characteristic equation
0 = det(I A(, )), A
ij
() L()
i
(e

e
j
)
(H) For = 0, the characteristic equation has a pair of simple roots
i
0
with
0
,= 0 and no other root that is an integer multiple of i
0
.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Hopf Bifurcation Theorem
If we express the characteristic equation as h(, ) = 0, then (H) implies
that the partial derivative h

(0, i
0
) ,= 0. = () = () +i() for
small satisfying (0) = i
0
. In particular, (0) = 0 and (0) =
0
.
Theorem
Let (H) and

(0) > 0 hold. Then there exists


0
> 0, real-valued even
functions () and T() > 0 satisfying (0) = 0 and T(0) = 2/
0
, and
a nonconstant T()-periodic function p(t, ), with all functions being
continuously dierentiable in for [[ <
0
, such that p(t, ) is a solution
of (24) and p(t, ) = q(t, ) where q(t, 0) is a 2/
0
-periodic solution
of q

= L(0)q.
Moreover, there exist
0
,
0
, > 0 such that if (24) has a nonconstant
periodic solution x(t) of period P for some satisfying [[ <
0
with
max
t
[x(t)[ <
0
and [P 2/
0
[ < , then = () and
x(t) = p(t +, ) for some [[ <
0
and some .
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Hopf Bifurcation Theorem
Theorem
(Conti.) If F is ve times continuously dierentiable then:
() =
1

2
+O(
4
)
T() =
2

0
[1 +
1

2
+O(
4
)]
If all other characteristic roots for = 0 have strictly negative real parts
except for i
0
then p(t, ) is asymptotically stable if
1
> 0 and
unstable if
1
< 0.

1
> 0:supercritical Hopf bifurcation.
1
< 0: subcritical Hopf
bifurcation.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Negative Feedback
Consider
x

(t) = f(x(t r))


f(0) = 0, f

(0) = 1, f

(0) = A, f

(0) = B
The equation exhibits negative feedback in view of the negative sign and
in the sense that xf(x) > 0, at least for x near zero. In the absence of
the delay, x = 0 would be an asymptotically stable equilibrium. By
s = t/r.
x(s) = rf(x(s 1))
The linearized equation about x = 0 is
v(s) = rv(s 1)
with the characteristic equation
+re

= 0
The roots have negative real part when 0 r < /2; = i/2 are
roots at r = /2 and all other roots have negative real part.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Negative Feedback
Set r = /2 +
F(, ) + (/2 +)e

= 0. (25)
F(i/2, 0) = 0, F

(i/2, 0) = 1 + (/2)i, F

(i/2, 0) = i.
the implicit function theorem implies that we can solve F = 0 for
= () = () +i() satisfying (0) = (/2)i and
d
d
(0) = F

/F

=
/2 +i
1 + (/2)
2
Consequently,
d
d
(0) =
/2
1 + (/2)
2
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Delayed Negative Feedback
When = 0 there exists > 0 such that the roots of the characteristic
equation consist of i(/2) and roots satisfying Re() < .
By the continuity of the roots with respect to , there exists
0
> 0 such
that for [[ <
0
, the roots of (25) consist of () i() and other
roots satisfy Re() < /2. Thus the hypotheses of the Hopf
bifurcation theorem are satised.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
The Logistic Eq.
N

(t) = N(t)[1 N(t r)]


Has positive equilibrium N = 1. Let
v = N 1
v

(t) = v(t r)(1 +v(t))


Let
X = log(1 +v)
x

(t) = f(x(t r))


f(x) = e
x
1
A = B = 1
The logistic equation has a supercritical Hopf bifurcation at r = /2
which is asymptotically stable.
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Second-Order Delayed Feedback System
The damped oscillator equation
x

+bx

+ax = F(t), a, b > 0


F(t) = F(x(t r))
xF(x) < 0, x ,= 0
F

(0) = d < 0.
x

(t) = v(t)
v

(t) = bv(t) ax(t) + F(x(t r))


Its only steady-state is x = 0 and the corresponding characteristic
equation is

2
+b +a = de
r
(26)
Let
X() = x(t), V () = v(t), r = t,

X() = rV ()

V () = brV () arX() +rF(X( 1))


Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Second-Order Delayed Feedback System
Lemma
There are no real roots satisfying 0
Lemma
There exists R > 0, depending only on a,b,d, such that if is a root with
Re() 0, then [[ < R.
Proposition
Let 0 r
1
< r
2
. Suppose that for r
1
r r
2
there are no roots of (26)
on the imaginary axis. Then
M(r
1
) = M(r
2
)
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
A Second-Order Delayed Feedback System
Consider = i where > 0.
a
2
+ib = de
ir
a
2
= d cos(r)
b = d sin(r)

4
+ (b
2
2a)
2
+a
2
d
2
= 0
Dene
= (b
2
2a)
2
4(a
2
d
2
)

=
1
2
[2a b
2

]
Then
< 0 implies no purely imaginary roots.
> 0, 2a b
2
< 0, a
2
> d
2
implies no purely imaginary roots.
> 0, 2a b
2
< 0, a
2
< d
2
implies one root i
+
.
> 0 2a b
2
> 0 a
2
< d
2
implies one root i
+
Applied Delayed Dierential Equations
Introduction Delayed Negative Feedback Existence of Solutions Linear Systems and Linearization Examples and Lyapunov Fu
Rich Dynamics in low-dim systems
Delayed two-neuron model, [Shayer and Campbell, 2000, SIAM AM]:
x
1
(t) = kx
1
(t) + tanh(x
1
(t
s
)) +a
12
tanh(x
2
(t
2
)),
x
2
(t) = kx
2
(t) + tanh(x
2
(t
s
)) +a
21
tanh(x
1
(t
1
)).
Delayed gene regulation networks, [Chen, Cheng, submitted]:
du(t)
dt
=
1
u(t) +
1
f
1
(v(t))
dv(t)
dt
=
2
v(t) +
2
f
2
(w(t))
dw(t)
dt
=
3
w(t) +
3
f
3
(u(t s)) +
4
f
4
(v(t r)),
Applied Delayed Dierential Equations

You might also like